Method and Arrangement for Route Cost Determination and Selection with Link Cost Interaction
This invention extends routing mechanisms that use link metrics for route selection so that: A link metric cross correlation vector is determined for all links, where each element in the vector corresponds to some other link, and reflects the change in the link metric value if a data flow would already use this other link. The invention further describes a specific embodiment where all crosscorrelating links are adjacent to each other, i.e., they terminate or originate in a common node. A mechanism is described to create an extended routing graph. This extended graph permits the use of standard polynomial time algorithms that simultaneously construct the optimal route and find the optimal route metric (such as shortestpath algorithms) also for the adjacent link crosscorrelating case.
The present invention relates to a method of routing traffic as defined in the preamble of claim 1 and to a node as defined in the preamble of claim 14.
BACKGROUND OF THE INVENTIONThe present invention applies to a communication network in which data can be transmitted from a sender to a receiver via a number of intermediate network nodes. Data can be transmitted via different routes, usually comprising one or more of the intermediate network nodes. A direct path between two adjacent nodes is referred to as a link. Depending on which communication links are used for a communication flow different transmission routes (or paths) through the network from the sender to the receiver can be distinguished.
It is important in any such communication system to optimize the selection of the routes to utilize the resources as efficiently as possible.
In existing routing solutions a route selection is typically based on assigning a cost value, or metric, to each communication route, that is, a route metric. The route metric is generally determined as the sum of the metric values for all links, or hops, in the route. The link metric values are often used to reflect the “quality” or utility of a particular link in relation to some performance objective. Typically, but not necessarily, each node determines link metric values for its own link interfaces and the values are distributed to other nodes using routing protocols. The best route is typically selected to be the route that has the best route metric.
In the prior art as described above each link is considered independently. The possible effect on a link of changes in any other link is not considered.
The “quality” or utility of a link can be affected by what happens in other links. This can be the case, e.g.,

 for different links sharing the same medium or channel, for example, the same CSMA medium used for transmission on different hops, with joined contention
 different links transmitting on different channels which interfere with each other, for example, different radio access technologies using the same or adjacent spectrum
 transmission of different links using common processing power, memory or other common resources, for example, if the links are within the same node and use the common DSP, RAM resources
In multiradio multihop networks, usually a number of different links exist to and from the communication nodes, the links using different radio access technologies or channels. In such a scenario, a routing decision where the route metric is only considered based on the individual link metrics leads to a nonoptimal solution, since the interfering effect of different links in a route is not considered.
OBJECT OF THE INVENTIONIt is an object of the present invention to enable more efficient use of network resources by improving routing through the network.
SUMMARY OF THE INVENTIONThis object is achieved according to the invention by a method as initially defined, having the characterizing features of claim 1 respectively. The object is also achieved according to the invention by a node as initially defined having the characterizing features set out in claim 14.
The basic idea underlying the invention is to reflect the interaction and coupling between the existing and potentially forthcoming links in the link metric, when determining a new route for data traffic. This invention improves the performance, service quality support, and capacity of communication networks by improving the routing decisions.
Briefly, the coupling between links relates to the physical layer including transmit power level and the path gain between a transmitter and an unintended receiver, but more importantly to the traffic load on a link as well as resource sharing constraints on, e.g., processing power or memory. In particular, a linear model of link metric interaction is proposed as one embodiment of the invention. A linear model (first order approximation) simplifies the problem and is accurate when the expected interaction change is relatively small in comparison to the link metric value. This is typically the case when traffic load is the dominating interaction factor.
This invention extends routing mechanisms that use link metrics for route selection so that:
A link metric cross correlation vector is determined for each link a, where an element in the vector corresponds to some other link b reflecting the change in the link metric value of a if a data flow would already use this other link b.
In a specific embodiment all crosscorrelating links (those with nonzero values in the cross correlation vector) are adjacent to each other, i.e., they terminate or originate in a common node.
A mechanism is described to create an extended routing graph. This extended graph permits the use of standard polynomial time algorithms that simultaneously construct the optimal route and find the optimal route metric (such as shortestpath algorithms) also for the adjacent link crosscorrelating case.
The idea is preferably realised primarily through extending routing mechanisms where each link has an associated link metric prior to route setup and the metric reflects the coupling between existing active links.
At route setup, a route is selected that minimizes a route metric, where the latter is composed of new updated link metrics along said route and the new updated link metrics is based on the old link metrics and the coupling between links.
Specifically, a link metric cross correlation vector is determined for each link a in addition to the metric value. Each element in this vector corresponds to some other link b. It is a measure of how the link metric value on a changes if the route selected for a data flow also uses link b.
Route selection mechanisms can use the link metric cross correlation vector to, e.g., avoid paths where the presence of a data flow in the preceding or following hops of two or more consecutive links unfavorably affects the link metric of the middle hop.
As the skilled person will readily understand, the invention can be applied to any routed communication network. However, the invention is particularly useful in wireless multihop networks with heterogeneous link technologies (e.g., radio access technologies). If two different link technologies are available between the same pair of nodes, then according to the invention they are treated as separate links with separate link metrics and link metric cross correlation vectors.
Note that this approach is particularly useful when combined with a predictive routing algorithm as described in copending application PCT/SE2005/000948, where the predicted metric after the admission of a new traffic flow to a path is considered for the routing decision.
In the following embodiments of the invention will be described in more detail, with reference to the appended drawings, in which
This invention extends link metric based routing mechanisms by introducing a link metric cross correlation vector σ_{a }for each link a in addition to the link metric m_{a}. Consider now the routing problem for a given data flow. Each component σ_{a}^{b }in the vector is a measure of how the link metric value m_{a }on link a changes if the route selected for the flow also uses link b. Note that some components in σ_{a }may be zero, which can mean that the corresponding link metrics are uncorrelated with m_{a }or that no correlation information is available. In practice σ_{a }may also be implemented as a set, where the elements are the pairs (b,σ_{a}^{b}) for which there is a correlation and/or information exists. The selfcorrelation σ_{a}^{a }is the selfinduced change in the link metric value when the data flow uses the link.
Estimated CorrelationCorrected Link and Route Metrics
The link metric cross correlation vector σ_{a }can be used by route selection mechanisms when computing route metrics and evaluating paths. A mechanism can, given a path p ε P where P is a set of candidate paths, compute estimated correlationcorrected link metrics
for all links a in p, where σ_{l}^{p}=(σ_{l}^{a}_{1}, . . . , σ_{l}^{a}_{p}) are the components in σ_{a }corresponding to the links b_{1}, . . . , b_{p} in a path p . The estimated correlationcorrected route metric can then be computed as appropriate, e.g., as the sum of the link metrics
In practice a linear model (first order approximation) can be used for the link metric interaction. The estimated correlationcorrected link metrics then can be computed as the sum of the link metric and the cross correlation values of the links in p:
Construction of Routes
The computation approach outlined above assumes that a set P of candidate paths p is formed first and then the estimated correlationcorrected route metric of each path p ε P is evaluated. In general this is necessary since each link can have a large number of possible estimated correlationcorrected link metric values (up to one value per candidate path) and the right one to choose is not decided until the path is selected. However, it is usually not feasible to enumerate all possible candidate paths in a network since the number of paths grows exponentially with the number of nodes/links.
In the following it will be discussed how to determine which paths are to be included in the set P of candidate paths.
One practical method for determining P is to:
 1. Ignore the link metric crosscorrelation vectors σ_{a }on all links a and use only the ordinary link metrics m_{a}.
 2. Use a kpath route construction method, e.g., a kshortest path algorithm to find the k shortest paths (the shortest, the second shortest, etc. up to the kth shortest) with respect to the link metrics m_{a }(see, e.g., Eppstein [1]).
 3. Let P be the k paths thus found.
This method will not find the optimal path with respect to the estimated correlationcorrected link metrics
In some cases it is possible to compute the optimal path with respect to
The general case where link metrics can correlate with any other link metric likely requires a centralized routing function. For the case with only adjacent link metric crosscorrelation, standard combinatorial route construction algorithms can be used. This admits at least a decentralized operation where each node obtains a network view from routing protocols that propagate link information. Here the link metric correlation vectors σ_{a }can be piggybacked on the link information, including the ordinary link metric values m_{a}, which is propagated by any existing routing protocol.
Determining the CrossCorrelation Values
There are a variety of methods for finding the correlation vectors σ_{a}:

 for links a which are terminated (or otherwise somehow controlled) at some central entity this correlation can be fairly exact based on detailed knowledge, e.g., about resource (such as timeslot, frequency, code, etc) to be scheduled, power level, etc
 for links a which are not controlled in a central way (links to/from different nodes) but which still interfere due to overlap in frequency/range/time a vaguer scheme of determining the correlation of metrics can be done, e.g., by observation over time how link metrics are correlated.
The latter case above can be accomplished by observing the link metrics before and after a routing case. Consider one link a on p. Let m_{a }and m′_{a }be the metric values of the link a before and after routing a flow on a path p. Without any a priori knowledge on how the link metric of this link correlates with the link metrics of the other links b in p, it is difficult to determine which link(s) induce the change from m_{a }to m′_{a }and how much this routing case represents the average crosscorrelation behavior. In general an updated estimate of the crosscorrelation value can be expressed as:
σ_{a}^{b}(new)=g(σ_{a}^{b},α_{a}^{b}h(m_{a}, m′_{a})), (3)
where g and h are some functions and α_{a}^{b }is a weight factor representing the probable influence of link b. The weight factor can, e.g., represent geographical distance if it is believed that links that are geographically close to each other are more likely to have correlating link metrics. The function h can be simply the difference between the link metrics and the function g could be a weighted combination or some moving average.
A problem with correlation vectors σ_{a }is that they depend on the link metrics m_{a}, i.e., it may be that if routes are changed for many paths during the routingdecision the correlation values will change as well. How the correlation values will change might be hard to estimate as these may have a nontrivial relation to the absolute link metrics (nonlinear) and in some cases be based on observations. This means that there might be practical limits on the amount of paths that can be considered in a single routing decision.
One approach to handling the dependencies between the link metrics and the cross correlation vectors is to combine the method with a predictive routing algorithm as described in the abovementioned copending application, where the predicted metric after the admission of a new traffic flow to a path is considered for the routing decision. In the terminology of the present invention the predictive routing algorithm can be viewed as estimating the selfcorrelation σ_{a}^{a}. The combined method naturally extends the prediction to all components of σ_{a}.
One or more of the nodes I, S, R may also comprise a path determining unit 7 for determining the set of possible paths, as discussed above. In addition one or more nodes I may comprise a crosscorrelation value determining unit 9 for updating the crosscorrelation values as discussed above. In
The link metric values before the routing decision for the four links are shown in Table 1. A12 denotes the path from node 1 to node 2 using technology A, A23 denotes the path from node 2 to node 3 using technology A, B12 denotes the path from node 1 to node 2 using technology B, B23 denotes the path from node 2 to node 3 using technology B,
The dashed line of
As can be seen in Table 2, in this example, the link metric values for the two links B12 and B23 have increased (for B23 significantly) as a result of the route selection. The reason for the increase can be that one of the links became very loaded following the routing, or that the two hops use the same link technology in this route and that they have to share resources or that they interfere with each other. A consequence is that the selected route was, after all, maybe not the best possible.
In
and
The routing mechanism can now use the link metric cross correlation vectors to determine the routing path. The dashed line again indicates the best route, which in this example is selected as the route with the smallest sum of the estimated correlationcorrected link metrics.
The path {B12, A23 } is clearly better than {B12, B23 } (in the sense of the route metric) since the actual link metric sum was reduced from 5,1 to 3,7.
This implies that the two paths from node S1 to node R1 are equally good (under this metric) and that they are equally likely to be used for an immediate routing case. It should be noted that this is a simplified expression that, for example, does not consider the case when f=c. In practice this case would be handled either by letting m be some (large) number when f=c or by replacing the capacity c in the expression by, e.g., c′=c+ε where ε is a small number.
Suppose now that links a and b use the same link technology and that they have to share time slots (links c and d use different link technologies and do not have correlated link metrics). Assume further that node 2 cannot receive and transmit simultaneously. Because of this there are situations when time slots are available on both link a and link b, yet they cannot be used together. One such situation is shown in the figure by the set of boxes next to links a and b. Here each box is a time slot and a shaded box is an occupied slot. The two links have free time slots available, but because of the simultaneous transmit/receive restriction there is no feasible combination of free timeslots in a and b. Suppose now that this situation occurs on average for every one hundredth routing case.
The link metric cross correlation vectors can be used to improve the routing decision, but the question is how to select the cross correlation values σ_{a}^{b }and σ_{b}^{a}. Assume that the conflicting scenario would be measured by saying that the link metric in this case is m=2c=12. Further assume that the average load f is three units. The expected metric in this case is
If additive link metric crosscorrelation values are used, then a reasonable estimate in this case can be to set σ_{a}^{b}=σ_{b}^{a}=0,1/2=0,5. This would prioritize the path over c and d under equal load situations, but still not preempt the use of the path over a and b completely.
Adjacent Link Cross Correlation—An Extended Routing Graph
A specific common case of this invention is when all crosscorrelating links are adjacent, that is, when, for a given link a=(i,j), the links b where σ_{a}^{b}≠0 all terminate at node i or originate at node j (b is either b=(k,i) or b=(j,k) for some other node k).
If single path routes are used and if the route metric admits polynomial time algorithms that simultaneously construct the optimal route and find the optimal route metric (such as shortestpath algorithms), then all of these algorithms can still be used to find the optimal path for the adjacent link cross correlation case.
In general this is not at all straightforward to do with the link metric cross correlation vectors since each link can have a large number of possible estimated link metric values (up to one value per candidate path). To construct the optimal route and compute the optimal route metric, a method is needed that picks out the right metric to use for each link. For the adjacent link cross correlation case then the following procedure can be used.
The idea is to first create an extended routing graph, then compute the shortest path in this graph, and finally to derive the route and route metric in the original graph. Note that links are directed (i.e. (i,j) is different from (j,i)) and that all metrics and estimated metrics are assumed to be nonnegative. Further, if two different link technologies (e.g., radio access technologies) are available between the same origin and destination nodes, then they are treated as separate links.
The extended routing graph is constructed by replacing each node by a number of different virtual nodes that represent the various roles that the node can have with respect to flows (a flow is here an endtoend routing case, i.e., a path between a source and a destination node).
Four separate cases of node roles need to be distinguished:
 1. The node is the final destination for a flow. In this case the metric on the incoming link is equivalent to the ordinary metric. This is true even if the incoming link metric is correlated with some outgoing link metric, since the outgoing link is not used by this flow. This situation is indicated by solid arrow 11 in
FIG. 4 .  2. The node is the origin for a flow. In this case the metric on the outgoing link is equivalent to the ordinary metric. The argument is similar to the one above. This situation is indicated by solid arrow 12 in
FIG. 4  3. The node is a transshipment node for a flow (i.e., not the origin or the destination), and neither the incoming nor the outgoing link metrics are correlated with each other. In this case the link metric on both the incoming and outgoing links is equivalent to the corresponding ordinary metrics. This situation is indicated by solid arrow 13 in
FIG. 4 .  4. The node is a transshipment node for a flow, and either the incoming link metric is correlated with the outgoing link metric or the outgoing link metric is correlated with the incoming link metric or both link metrics are correlated with each other. In this case the link metric on the incoming and outgoing links need to reflect the corresponding crosscorrelation values. This situation is indicated by solid arrow 14 in
FIG. 4 .
The four cases are shown in
Cases 1, 2, and 3 only require one virtual node each—cases 1 and 2 apply to all flows where the node is the destination or the origin, and for case 3 the same virtual node can be used for all pairs of incoming and outgoing links with uncorrelated link metrics. However, case 4 requires one virtual node for each pair of incoming and outgoing link where either (or both) of the link metrics are correlated with the other.
Formally the extended routing graph is constructed as follows:
 1. Take a node i in the original graph that has not been considered yet and add:
 a) one destination virtual node i^{t},
 b) one origin virtual node i^{s},
 c) one nocorrelation transshipment virtual node i_{0,0 }provided that at least one pair of incoming and outgoing links exists where the link metrics are uncorrelated, i.e., if
 ∃a,b:a=(k,i),b=(i,j) where σ_{a}^{b}=0 and σ_{b}^{a}=0, and finally
 d) one correlation transshipment virtual node i_{a,b }for each pair of incoming link a and outgoing link b where either (or both) of the corresponding link metrics are correlated with the other, i.e.,
 ∀a,b:a=(k,i),b=(i,j) where σ_{a}^{b}≠0 or/and σ_{b}^{a}≠0.
 2. If all nodes in the original graph have been considered then go to 3, otherwise go to 1.
 3. Next take a link a=(i,j) in the original graph that has not been considered yet and add:
 a) one virtual link (i^{s},j^{t}) with link metric
m =m_{a},  b) one virtual link (i^{s},j_{0,0}) with link metric
m =m_{a},  c) virtual links (i^{s},j_{a,b}) with link metrics
m f_{a}(m_{a},σ_{a}^{b}) for all b where j_{a,b }exists,  d) one virtual link (i_{0,0},j^{t}) with link metric
m =m_{a},  e) one virtual link (i_{0,0},j_{0,0}) with link metric
m =m_{a},  f) virtual links (i_{0,0},j_{a,b}) with link metrics
m =f_{a}(m_{a},σ_{a}^{b}) for all b where j_{a,b }exists,  g) virtual links (i_{b,a},j^{t}) with link metrics
m =f_{a}(m_{a},σ_{a}^{b}) for all b where i_{b,a }exists,  h) virtual links (i_{b,a},j_{0,0}) with link metrics
m =f_{a}(m_{a},σ_{a}^{b}) for all b where i_{b,a }exists, and  i) virtual links (i_{b,a},j_{a,c}) with link metrics
m =f_{a}(m_{a},σ_{a}^{b},σ_{a}^{c}) for all b and c where i_{b,a }and j_{a,c }exists.
 a) one virtual link (i^{s},j^{t}) with link metric
 4. If all links in the original graph has been considered then terminate, otherwise go to 3.
Note that some of the cross correlation values in the link metrics in substeps 3c, 3f, 3g, 3h, and 3i may be zero. This happens when the involved correlation transshipment virtual nodes have been added to reflect the correlation of other adjacent links. However, the virtual links are still required in order to maintain the original graph's connectivity.
The procedure described above always adds destination and origin virtual nodes, but if it is known that a node will never be origin and/or destination in any routing case then the corresponding virtual nodes and links can of course be removed.
Examples of the extended routing graph construction is shown in
The extended routing graph now permits any standard combinatorial route construction algorithm, e.g., a shortestpath method, to find the optimal route and total route metric for a given flow. If the origin and destination nodes of the flow are o and d respectively, then the virtual nodes o^{s }and d^{t }should be used in the extended routing graph. The optimal route metric is found directly from the extended routing graph, and the route in the original graph is simply constructed by substituting each virtual node by its corresponding original node and similarly for the virtual links.
As can be seen in
The metrics for the paths comprising a single link, or hop, are shown in Table 4:
For the path between sending node N1 and destination node N3 via node N2, the link metrics are as follows:
From node N1 to node N2: m_{a }
From node N2 to node N3: m_{d}+σ_{d}^{s }
For the path between sending node N1 and destination node N4 via node N2, the link metrics are as follows:
From node N1 to node N2: m_{a}+σ_{a}^{b }
From node N2 to node N4: m_{b }
For the path between sending node N1 and destination node N4 via node N3, the link metrics are as follows:
From node N1 to node N3: m_{c }
From node N3 to node N4: m_{e}+σ_{e}^{c }
For the path between sending node N1 and destination node N4 via both node N2 and N3, the link metrics are as follows:
From node N1 to node N2: m_{a }
From node N2 to node N3: m_{d}+σ_{d}^{a}+σ_{d}^{e }
From node N3 to node N4: m_{e }
For the path between sending node N2 and destination node N4 via node N3, the link metrics are as follows:
From node N2 to node N3: m_{d}+σ_{d}^{e }
From node N3 to node N4: m_{e }
To determine the path metric for a particular path, the link metrics for all links in that path are added. The path having the lowest path metric is generally the one that should be selected.
In step S1 the set P of candidate paths p is determined.
In step S2 a path p in the set P is selected, for which the metrics should be determined.
In step S3 the correlationcorrected link metric value for each link in the path p is calculated. This calculation is based on the crosscorrelation vector for the respective link and the current link metric, as explained above.
In step S4 it is determined if the correlationcorrected link metric values have been calculated for all paths. If no, return to step S2 and select a path p for which the link metric values have not been calculated; if yes, go to step S5.
Step S5: The optimal path for the connection is selected, based on the correlationcorrected link metric values for the links in each path. In practice this may involve, for each path, adding the corrected link metric values for all links in the path. The resulting values for all paths can than be compared and the path having the most favourable metric value can be selected.
In an optional step S6 traffic is transmitted on the selected path.
How to obtain the crosscorrelation values used in step S3 is discussed above.
Step S10: Create the extended routing graph by introducing virtual nodes and virtual links as described above.
Step S11: Determine the shortest path by applying a standard shortest path algorithm to the routing graph created in step S10.
Step S12: Extract the route and the route metric. The route is found by replacing each virtual node by its corresponding node and each virtual link by its corresponding link. The route metric is identical to the route metric in the extended routing graph.
Note that some metrics may be better when it has a large value (for example, supported rate, link quality, etc) while other metric may be better when it has a small value (for example, delay, number of hops, etc). Different metrics can affect each other. One solution for combining the different metrics into one comparable value can be using a utility function, known per se.
REFERENCES
 [1]. Eppstein, D., “Finding the k Shortest Paths,” SIAM Journal on Computing, Volume 28, Number 2, 1998.
Claims
126. (canceled)
27. A method of selecting a route for a data transmission flow in a communication network, comprising:
 defining a first route from a sender to a receiver comprising at least a first link having a first link metric, and a second route from the sender to the receiver comprising at least a second link having a second link metric;
 determining the value of at least the first link metric based on a crosscorrelation value representing the change that would occur in the link metric value if a data flow would already use another link in the network; and
 comparing route metrics for the first and second routes, the route metrics based on the link metric values of the links comprising the routes.
28. The method of claim 27 wherein the first route comprises at least two links and the other link considered in the crosscorrelation value is included in the first route.
29. The method of claim 27 wherein the first link metric value is calculated on the basis of a link metric crosscorrelation vector comprising, for the first link, a crosscorrelation value corresponding to another link, and reflecting the change in the link metric value for the first link if a data flow would already use the other link.
30. The method of claim 27 wherein the first link metric value is calculated on the basis of a set of values comprising pairs (b,σab) for which there is a correlation and/or information exists, b being another link and (σab) being a crosscorrelation value reflecting the change in the first link metric value if a data flow would already use the other link.
31. The method of claim 30 wherein the first link metric value is calculated according to the function ma=fa(ma,σab), where ma is the first link metric value to be calculated, ma is the current first link metric value and σab is the crosscorrelation value reflecting the change in the first link metric value if a data flow would already use the other link b.
32. The method of claim 31 wherein the first link metric value is calculated according to the function m a _ = m a + ∑ b ∈ p σ a b, where ma is the first link metric value to be calculated, ma is the current first link metric value, P is the set of all links to be considered and σab is a crosscorrelation value reflecting the change in the first link metric value if a data flow would already use the other link b.
33. The method of claim 27 further comprising calculating the first link metric value based on the change that would occur in the first link metric value if a data flow would already use the first link.
34. The method of claim 27 wherein defining a first route from a sender to a receiver comprises:
 determining a set P of possible paths from a sender to a receiver by using link metrics ma associated with links and a standard kshortest path construction method to find a predetermined number k of shortest paths;
 including said k shortest paths in the set P; and
 selecting a first route from the set P.
35. The method of claim 27 wherein the crosscorrelation value is determined based on knowledge about how links affect each other.
36. The method of claim 27 wherein the crosscorrelation value is determined based on observations over time of how link metrics are correlated.
37. The method of claim 27 further comprising calculating a corrected crosscorrelation value σab(new) as:
 σab(new)=g(σab,αabh(ma,ma))
 where g and h are functions, αab is a weight factor representing the probable influence of the link b on the link a, ma is the current link metric value for the link a, and m′a is the link metric value for the link a when a flow uses a path including the link a.
38. The method of claim 27 further comprising:
 creating an extended routing graph by introducing virtual nodes and virtual links;
 determining the shortest path by applying a standard shortest path algorithm to the extended routing graph;
 extracting a route along the determined shortest path by replacing each virtual node by its corresponding node and each virtual link by its corresponding link; and
 extracting a route metric from a corresponding route metric in the extended routing graph.
39. A telecommunications network node, comprising:
 a link metric function unit (LMF) operative to monitor or control a link characteristic needed for determining a link metric for a first link starting at the node, the LMF further operative to calculate a link metric value of the first link on the basis of a link metric crosscorrelation vector comprising a crosscorrelation value corresponding to another link and reflecting the change in the first link metric value if a data flow would already use the other link.
40. The node of claim 39 wherein the LMF is operative to calculate the link metric value on the basis of a set of values comprising pairs (b,σab) for which there is a correlation and/or information exists, b being another link and (σab) being a crosscorrelation value reflecting the change in the first link metric value if a data flow would already use the other link.
41. The node of claim 39 wherein the LMF is operative to calculate the first link metric value according to the function ma=fa(ma,σab), where ma is the first link metric value to be determined, ma is the current first link metric value and σab is the crosscorrelation value reflecting the change in the first link metric value if a data flow would already use the other link b.
42. The node of claim 41 wherein the LMF is operative to calculate the first link metric value according to the function m a _ = m a + ∑ b ∈ p σ a b, where ma is the first link metric value to be calculated, ma is the current first link metric value, P is the set of all links to be considered and σab is a crosscorrelation value reflecting the change in the first link metric value if a data flow would already use the other link b.
43. The node of claim 39 wherein the LMF is operative to calculate the link metric value of the first link based on the change that would occur in the first link metric value if a data flow would already use the first link.
44. The node of claim 39 wherein the LMF is operative to determine a set P of possible paths by:
 using link metrics ma associated with links and a standard kshortest path construction method to find a predetermined number k of shortest paths; and
 including said k shortest paths in the set P.
45. The node of claim 39 wherein the LMF is operative to use a crosscorrelation value determined based on knowledge about how the links affect each other.
46. The node of claim 39 wherein the LMF is operative to use a crosscorrelation value determined based on observations over time of how link metrics are correlated.
47. The node of claim 39 wherein the LMF is operative to calculate a corrected crosscorrelation value σab(new) as follows:
 σab(new)=g(σab,αabh(ma,m′a))
 where g and h are functions, αab is a weight factor representing the probable influence of the link b on the link a, ma is the current link metric value for the link a, and m′a is the link metric value for the link a when a flow uses a path including the link a.
48. The node of claim 39 wherein the LMF is operative to:
 create an extended routing graph by introducing virtual nodes and virtual links;
 determine the shortest path by applying a standard shortest path algorithm to the extended routing graph;
 extract a route along the determined shortest path by replacing each virtual node by its corresponding node and each virtual link by its corresponding link; and
 extract a route metric from a corresponding route metric in the extended routing graph.
49. A communications network comprising at least one node according to claim 39.
Type: Application
Filed: Dec 1, 2005
Publication Date: Jul 9, 2009
Patent Grant number: 9674073
Inventors: Mikael Prytz (Rönninge), Joachim Sachs (Aachen), Per Magnusson (Linköping), Johan Lundsjö (Spänga), Peter Larsson (Solna)
Application Number: 11/917,112
International Classification: H04L 12/26 (20060101);